A simple score to predict early death in adult cancer patients with bloodstream infections.

OBJECTIVE:To develop a simple score to predict early death in adult cancer patients with bloodstream infection. METHODS:A prediction score was developed by analyzing data collected from 361 episodes of BSI at an oncology cancer care center in Brazil. Early death was defined as all deaths occurring within 7 days of the first positive blood culture. RESULTS:Multivariate regression analysis identified poor Karnofsky performance status, uncontrolled cancer, hypotension, pulmonary infiltrates, associated infectious sites, and neutropenia as independently associated with death. Predictive scores were developed assigning points to each significant independent variable. The overall mortality was 20.5% and, the total weighted score ranged from 0 to 11 points, with a very good calibration (Hosmer-Lemeshow statistic, P=0.92) and discrimination (area under receiver operating characteristic curve=0.893). The cutoff value for the predictive score was 3 points, with a negative predictive value of 99% and sensitivity of 98.6%. CONCLUSION:The score model was able to identify adult cancer patients with bloodstream infection at lower risk for early death with an elevated degree of certainty as depicted by a very high negative predictive value. It is essential to prospectively validate the rule in a different group of patients.